Probably Anytime-Safe Stochastic Combinatorial Semi-Bandits

Watch a 58-minute lecture by Prof. Vincent Y. F. Tan from the National University of Singapore as he presents groundbreaking research on stochastic combinatorial semi-bandits with safety constraints. Explore a novel approach to online decision-making that addresses risk management when selecting multiple items simultaneously. Learn about the PASCombUCB algorithm, which optimizes regret while maintaining probabilistic safety guarantees across time horizons. Understand the theoretical foundations, including problem-dependent and problem-independent paradigms, and see practical applications in recommendation systems and transportation. Delve into topics such as permissible upper bounds, safe solutions, exposure consistency, and experimental results that demonstrate the algorithm's effectiveness. Benefit from the expertise of Prof. Tan, a distinguished academic whose contributions to information theory, machine learning, and statistical signal processing have earned him numerous accolades, including the MIT EECS Jin-Au Kong Outstanding Doctoral Thesis Prize and the Singapore National Research Foundation Fellowship.